Deep-Learning vs Classical Machine-Learning comparison for text classification

Abstract

The Classical Machine Learning and Deep Learning models are used to provide solutions in everyday technologies, like weather prediction, stock price prediction, voice-to-text conversion, fraud detection, quality assurance, etc. These implementations are only a part of a broad range of applications where these algorithms can offer unique services. In this dissertation, Classical Machine Learning models will be compared with Deep Learning Neural Network models, within the frame of Text Classification. This comparison will be done by using three different feature selection metrics, namely tf.idf, chi square (x2) and devmax.tf. Also, different Neural Network Deep Learning architectures are tested and compared between them, as well as different parameters (input vector size, topology architecture, etc.), which are applied in Neural Networks.